@HPC Podcast Archives - OrionX.net - HPC News Bytes – 20250127

Episode Date: January 27, 2025

- Massive AI datacenter investments: $500B Stargate, $600B Saudi - When datacenters can’t get bigger - India enters the chip manufacturing market [audio mp3="https://orionx.net/wp-content/uploads/2025/01/HPCNB_20250127.mp3"][/audio] The post HPC News Bytes – 20250127 appeared first on OrionX.net.

Welcome to HPC News Bites, a weekly show about important news in the world of supercomputing, AI, and other advanced technologies. Hi, everyone. Welcome to HPC News Bites. I'm Doug Black of Inside HPC, and with me is Shaheen Khan of OrionX.net. Shaheen, this past week felt as though we're living in what I would call real-time historical whitewater, to coin a phrase. This is an effect Donald Trump has on events, his outsized personality, the intense media glare he attracts, his willingness to take on big and risky

promises. All of this was on display in his first week back in the White House. And we're seeing how AI has become one of the biggest political and geopolitical issues of our time. On his second day in power, Trump announced what he billed as a $500 billion investment in AI data center infrastructure in the U.S. And alongside him were three of the biggest technology industry players, Sam Altman of OpenAI, Larry Ellison of Oracle, and Masayoshi Son of SoftBank. Later in the week, Saudi Arabia said they're interested in investing $600 billion in American AI. That's over a trillion dollars in US AI if it all comes together. And that has major implications in a lot of different ways.

You said, if it all comes together, you and Musk both. Okay. I guess we are participating in the celebrity ball here. Musk tweeted that the group doesn't have anything close to the first installment of that money. Zuckerberg said they're investing $65 billion just this year and just for their own internal use. And Microsoft said it doesn't know about who else has how much money, but it's good for the $80 billion that they announced. So yes, the entertaining part carries on. It's important to invest for the long term. So saying such and such amount of money over five years or 10 years has the reassuring aspect that it can sustain the effort towards actual progress. but it skips over

the immediate investment that is necessary, which is what Musk is referencing. It also can make it sound great while reporters are covering, but then fail to materialize and fade in the background as everyone forgets it. I don't expect that to be the case here since Ellison, Holtman, and Massa-san all have serious interest to make it happen, and partnering with the government helps achieve that. So we'll see how all of this unfolds, but what is true is the steadily rising global strength of the U.S. in many areas of technology, and AI is the big prize. Now, AI might have temporary setbacks, but the big trend is not about to stop, and that

has meant this kind of massive monies for infrastructure build-out. AI needs data, models, algorithms, and infrastructure. The big change over the past 15 years has been the growing importance of infrastructure. GPUs enabled neural nets to go deep, and that turned out to work surprisingly well. And more computational power has steadily overcome limitations of AI. It looks like intelligence is computable for real, but needs a lot of computational power. And that has fueled a stampede to build bigger data centers. But just like there is a Moore's law wall, there will be a limit to how big a data center can get and still be effective. This is another

place where HPC can lead the way. We've covered the strategies we see among exascale sites and their scientific instruments to move towards tighter integration of those sites into a logical single site. Well, that's where commercial sites seem to have to go as well, all enabled by faster wide area networks and tighter orchestration of workflow and data flow. But AI is still evolving, and besides deep learning and generative AI, more breakthroughs are needed and expected. So if the focus is solely on AI, one has to keep a cautious eye on the impact of better quality data, new models, faster algorithms, and new hardware technologies that might reduce the need for massive data centers. What you don't want is to invest a lot

of money in something that might suddenly go out of favor. A geographically distributed but logically centralized supercomputer is a way to help reduce that risk also. Yes, it feels as though we're not just in a race, but in an all-out, lung-bursting, and endless foot race to build out HPC AI data center capacity. On that theme, as you said, Beta announced last week it plans an investment of as much as $65 billion on data center expansion in 2025 alone. And to paraphrase Nelson Rockefeller, tens of billions here, tens of billions dollars there, pretty soon you're talking about serious money. These ideas were reported in an article in The Register,

and it does remind us of DOE's strategy around the IRI, the integrated research infrastructure, which will be a big part of the next generation of leadership supercomputing systems for the U.S. National Labs and other research institutions. And it will link up advanced computing resources from around the country. The era of the standalone leadership class supercomputer is coming to an end. All of this is happening in the face of a power grid already under strain from existing demands. So as we watch the installation of increasingly large-scale high-performance data center technology,

we also wait for a way to address the big limiting factor in all of this, not enough electrical power. With the growing importance of chips in our digital world, there's been significant push by many countries to manufacture processors within their own borders. The benefits are typically listed as supply chain resilience, transformative economic models, advancing skill sets, and the like. There's also a lot of money to be made. We hear countries as varied as Mexico and the Dominican Republic want to get into chip fabs in a big way. The Indian diaspora is very prominent in the chip world, but local manufacturing in India has been notably absent.

Well, this week there was news that the first made-in-India chips are expected to roll out this summer using 28 nanometer lithography. Yeah, India is a beneficiary of geotechno-politics in a fragmenting world. Its readiness to participate and lead with advantageous demographics and educated labor force and global and generally harmonious presence. India is well positioned to be an alternative to China and Korea. So there's a lot going on in India in high tech, driven by government ministries, an initiative formulated as Digital India Corporation, and business units under it. India Semiconductor Mission is one such business unit, reportedly with some autonomy and focused on long-term strategies for chips and display manufacturing and associated design and R&D.

Stronger ties with the U.S. and other Western and R&D. Stronger ties with the US and other Western governments are also critical. They have led to more foreign direct investments, and it all helps transform local companies as well. In the chip world, there's a lot of R&D and design in India, but not so much manufacturing. Micron has been mentioned as investing to build factories there too. There are large markets for 28 nanometer technology, so that's a great start and can be a foundation for more advanced chips and building the local supply chain. It's going to be a multi-decade effort in a very fertile ground, and we should expect to see more visible presence in HPC from India as well.

All right, that's it for this episode. Thank you all for being with us. HPC Newsbytes is a production of OrionX in association with InsideHPC. Shaheen Khan and Doug Black host the show. Every episode is featured on InsideHPC.com and posted on OrionX.net. Thank you for listening.